On this page
- The headline numbers
- Cite this page
- Q4 2026 rounds so far
- November 2026 rounds
- December 2026 rounds
- Acquisitions and shutdowns
- Three rounds that close out Q3
- Q3 2026 final close
- AI agent funding history
- What the pattern says
- What to watch in Q4
- Methodology
- Why Gravity has not raised
- FAQ
- Update log
- Previous editions
- Sources
The headline numbers
Derived from the round tables on this page, recomputed automatically every day. Each line is written to be quoted as it stands; please keep its date.
- Rounds so far
- As of 13 November 2026, Q4 2026 has 18 disclosed AI agent funding rounds worth $1,656.0 million in disclosed money.
- Largest round
- Manus's $500.0 million Not stated round on 8 October 2026 is the largest of Q4 2026 so far, 30% of the quarter's disclosed total as of 13 November 2026.
- Pace
- Q4 2026 is running at 0.41 rounds a day over its first 44 days (to 13 November 2026), against 0.64 a day across all of Q3 2026 (59 rounds in 92 days).
- Same day in Q3 2026
- At the same point of Q3 2026 (day 44, 13 August 2026) this tracker had 27 rounds worth $1,582.3 million, so Q4 2026 is 9 rounds behind and $73.7 million ahead on disclosed money as of 13 November 2026.
- Median and average
- The median disclosed Q4 2026 round is $44.8 million and the average is $92.0 million as of 13 November 2026. Q3 2026 closed at a $30.0 million median and a $107.1 million average ($74.5 million without Cognition's $2,000.0 million floor).
- Top category
- Consumer agents leads Q4 2026: 1 of 18 tagged rounds and 30% of disclosed money as of 13 November 2026.
Updated . Data through 13 November 2026. Every Q4 2026 row: data.csv (CC BY 4.0).
Data cut-off: 12 November 2026. Q4 is open and checked weekly, and the total above is a floor that will move. Every total on this page carries the date its research window closed, so a reader arriving in December can tell a current figure from a stale one without emailing me.
This is the running record of disclosed AI agent startup funding rounds in Q4 2026. A small count stays a small count, and the opening of a quarter stays an opening here. Precision about what is known, and about where the record stops, is the reason anyone cites this page. The substance of this first edition sits further down: the Q3 close moved, and it moved a lot.
The Q3 2026 funding tracker stays the per-round record for July, August and September, with a named source on every line. This page carries Q4 forward and corrects the Q3 total.
- Q4 2026 stands at $872.5 million across five rounds, data through 16 October 2026. Manus's $500 million floor and Armadin's $255.5 million Series B are 87 percent of that, so the quarter's early shape is two very large rounds and three much smaller ones.
- Q3 2026's final close is 59 rounds and about $6.32 billion, up from the 56 rounds and about $5.25 billion the Q3 page recorded on 28 September. Three rounds dated 28 to 30 September account for the difference.
- Instinct's $1.0 billion at a $10 billion valuation is the largest single addition, a Series C raised weeks after the company launched its invite-only consumer agent in August 2026.
- Agent security now reads like its own funding category. Reco ($55 million, 29 September) and Armadin ($255.5 million, 1 October) are consecutive rounds about securing or attacking agents rather than building them.
- Exclude Cognition and Q3 is about $4.32 billion. A single round recorded at a $2 billion floor was roughly a third of the quarter, which makes the unadjusted total a bad input for any quarter-over-quarter comparison, so the 58-round basis is the one to reach for when you set Q3 against another quarter. Whichever basis you pick, say which one it was, because the two bases differ by exactly $2 billion on the same quarter.
- No Q4 forecast appears on this page: no projection and no annualised run rate. The pace lines in the headline numbers compare Q4 with Q3 at the same day count, which is a description of what has happened so far, not a prediction.
Q4 2026 rounds so far
18 disclosed rounds, $1,656.0 million, dated 1 October 2026 to 13 November 2026. The table below is open and gets checked weekly for the rest of the quarter.
| Company | Amount | Stage | Valuation | Date | Category | What it does |
|---|---|---|---|---|---|---|
| Armadin | $255.5M | Series B | More than $2.5B | 1 Oct 2026 | Security | Offensive security. Chains individual security weaknesses into actionable attack paths so a defender can test them. Founded by one of Mandiant's founders. |
| Photon | $4.5M | Seed | Not disclosed | 1 Oct 2026 | Agent infrastructure | Tools and an API for developers who put AI agents inside messaging apps, including iMessage and WhatsApp. Tech Funding News describes the company as Vercel-backed. |
| Vocca | $20M | Series A | Not disclosed | 6 Oct 2026 | Vertical agents | AI agents for medical practices that handle patient calls, appointment scheduling, exam-preparation questions and administrative requests. Norrsken VC led, with Heal Capital, Speedinvest and Firstminute Capital. Total raised $25M. |
| Stuut | $52.5M | Series B | Not disclosed | 7 Oct 2026 | Vertical agents | AI platform that runs order-to-cash for finance teams: collections, cash application, payments, disputes and deductions. Insight Partners led; Andreessen Horowitz and M12 took part. Total raised $93M. |
| Manus | $500M | Not stated | Reported $4B, not confirmed | 8 Oct 2026 | Consumer agents | General-purpose AI agent that completes multistep tasks on a user's computer. Recorded at a $500M floor because the report says "more than $500 million". Boyu Capital led, with Tencent. The $4B valuation is Bloomberg's September report, not confirmed by the company. |
| Hone | $60M | Seed | $285M | 8 Oct 2026 | Agent platforms | Builds agents ("Engines") that take ongoing responsibility for business outcomes over weeks or months. Benchmark and Index Ventures led. No revenue or customer outcomes disclosed. |
| Town | $108M | Series B | $1B reported | 19 Oct 2026 | Agent platforms | Gives each employee an AI assistant ("Townies") that can now ask other assistants questions and hand off tasks across a team. Forerunner Ventures and Menlo Ventures co-led. |
| Daytona | $60M | Series B | $1B | 20 Oct 2026 | Agent infrastructure | Isolated cloud sandboxes where AI agents run code and operate software. FirstMark Growth led. Total raised $91M. The company says revenue run rate is up about 25x since January, from an undisclosed base. |
| Asymptote Labs | $5.5M | Seed | Not disclosed | 20 Oct 2026 | Security | Runtime visibility and policy enforcement for AI agents, with an open-source telemetry layer for agent activity. Tower Research Ventures led. |
| Omnichat | $10M | Series A | Not disclosed | 27 Oct 2026 | Vertical agents | Marketing, sales and customer-service agents that run inside WhatsApp, LINE, Instagram and other chat channels for retail brands. Openspace Capital led, with Darwin Ventures and Acorn Pacific Ventures. |
Q4 2026 to date: 18 rounds, $1,656.0 million. Data through 13 November 2026. Open quarter, checked weekly.
Armadin's investors are now confirmed. TechCrunch's 1 October report says Andreessen Horowitz and Accel led the Series B at a valuation of more than $2.5 billion, with Bain Capital Ventures, Redpoint, 8VC, Ballistic Ventures, GV, In-Q-Tel, Kleiner Perkins and Menlo Ventures joining. The 4 October edition left the leads out because only a search summary had them; the report itself settles it, so the valuation column now carries the $2.5 billion floor. Photon's investors are not recorded here.
Armadin is worth one more line because of its size. A $255.5 million Series B at the start of a quarter is not a seed-stage bet on a category forming. It is a growth cheque into agent-adjacent security, and it arrived one day after Reco's $55 million growth round in the same space.
Three rounds joined in the 16 October pass: Stuut's $52.5 million Series B (Insight Partners), Manus's round of more than $500 million (Boyu Capital, with Tencent) and Hone's $60 million seed (Benchmark and Index Ventures). Manus carries a reported $4 billion valuation from Bloomberg's September report that the company has not confirmed, so the valuation column says so and the amount sits at the $500 million floor.
The 30 October pass adds five rounds: Vocca's $20 million Series A (6 October, missed in the earlier passes), Town's $108 million Series B, Daytona's $60 million Series B for agent sandboxes, Asymptote Labs' $5.5 million seed for agent runtime security, and Omnichat's $10 million Series A. Town and Daytona both carry a reported $1 billion valuation, which sits in its own column and never enters a total. Two things were left out: Collibra's purchase of trail ML, because the target sells AI governance software rather than an agent product, and Modal's mid-October round, because I could not tie it to an agent-specific product.
Q4 2026 now has 18 disclosed rounds, so the headline numbers carry its median and average cheque size. The Q3 record had 59 rounds to average across.
November 2026 rounds
| Company | Amount | Stage | Valuation | Date | Category | What it does |
|---|---|---|---|---|---|---|
| Finpilot | $10M | Series A | Not disclosed | 4 Nov 2026 | Vertical agents | AI agents that take paperwork off investment teams at hospitals, pension funds and foundations. Lobby Capital led, with Madrona. Total raised about $14.5M. |
| firmly | $15M | Series A | Not disclosed | 6 Nov 2026 | Vertical agents | Agentic commerce platform that connects merchants to AI shopping channels; launched Ask Agent. Squarepoint Capital led, with J.P. Morgan Payments, Mastercard and Samsung NEXT. Borderline on scope, included on the same basis as Omnichat. |
| Bay | $37M | Series A | Not disclosed | 10 Nov 2026 | Security | Reviews an AI agent's proposed actions against its whole session before they run, then allows, asks for approval, redirects or blocks. Battery Ventures led the Series A. Also discloses a previously unannounced $13M seed (YL Ventures and Striker); $50M in total. |
| Daylight Security | $80M | Series B | Not disclosed | 10 Nov 2026 | Security | Managed detection and response in which AI agents handle routine security work and engineers handle judgment. Growth Equity at Goldman Sachs Alternatives led; Bain Capital Ventures and Craft Ventures returned. Borderline on scope (a managed service built on agents), included on the same basis as firmly. |
| Lumbra | $25M | Series A | Not disclosed | 10 Nov 2026 | Agent platforms | Model-neutral operating system for AI agents in high-consequence environments, which the company says already runs on top-secret government networks. Point72 Ventures and Ansa Capital co-led, with Costanoa. |
| Rox | $100M | Series B | $1.3B | 11 Nov 2026 | Vertical agents | AI agents for sales teams that research accounts, find prospects, prepare outreach and manage deals, positioned against traditional CRM. General Catalyst and March Capital co-led. Previously raised $50M (seed and Series A, November 2024). |
| Plurai | $13M | Seed | Not disclosed | 11 Nov 2026 | Agent infrastructure | Simulation infrastructure that finds and fixes AI agent failures before production; the platform reached general availability the same day. Team8 led, with U&I, Velocity and Mercer. |
| LangChain | $300M | Series C | $3B | 13 Nov 2026 | Agent platforms | Developer platform for building, evaluating, deploying and governing AI agents (LangSmith). CapitalG led, with GIC, DFJ Growth, IVP and Sapphire Ventures. |
November 2026 to date: 4 rounds, $162.0 million, data through 12 November 2026. Held back for want of a primary source or a clear scope fit: a reported $35M Graphon Labs Series A, Sentra's $150M (data security), Nebuly's $21M (measures agents rather than building them) and Shackleton's $5M seed.
December 2026 rounds
Also empty by design, for the same reason. The December table arrives at the December update and closes the quarter. When it lands, this page will carry a Q4 2026 final close with its own shown arithmetic, in the same format as the Q3 close below.
Acquisitions and shutdowns this quarter
Money leaves the category as well as entering it, and nobody keeps a running list of the exits for agent companies specifically. This section does: every acquisition, shutdown or pivot of a company whose product is an AI agent, an agent platform, agent infrastructure or an agent-building tool, each with the announcement it came from.
As of 13 November 2026, Q4 2026 has 5 acquisitions of AI agent companies, 0 shutdowns and 0 pivots on record, with $325.0 million in disclosed deal value. The table also carries 2 deals from the last 21 days of the previous quarter, marked with the quarter, because they landed after that edition closed.
| Company | What happened | Amount | Date | Detail and source |
|---|---|---|---|---|
| Listen Labs | Acquired by Salesforce | Undisclosed | 29 Sep 2026 (Q3) | AI customer research platform whose agents design studies, interview participants and synthesise the findings. Definitive agreement signed, expected to close in Salesforce's fiscal Q4 2027. Salesforce did not disclose terms; MarTech reports that many outlets put the price at about $2 billion. Source: Salesforce, 29 Sep 2026; MarTech, 30 Sep 2026. |
| DoubleO.ai | Acquired by ZoomInfo | Undisclosed | 30 Sep 2026 (Q3) | Agentic orchestration platform, founded in 2024, built to make multi-agent go-to-market automation reliable. ZoomInfo launched Agent Teams on its architecture the same day. Terms not disclosed. Source: ZoomInfo via Business Wire, 30 Sep 2026; Silicon Florist, 1 Oct 2026. |
| Ultravox | Acquired by Inworld | Undisclosed | 1 Oct 2026 | Platform developers use to build real-time voice agents. Ultravox staff who joined Inworld keep developing the platform, and its built-in Inworld voices move to Inworld's Realtime TTS-2 model. Terms not disclosed. Source: Inworld via Business Wire (Yahoo Finance), 1 Oct 2026. |
| Boost.ai | Acquired by Automation Anywhere | Undisclosed | 7 Oct 2026 | Norwegian conversational and voice AI platform for regulated industries such as banking, sold by Nordic Capital. Definitive agreement; expected to close in Q4 2026 subject to regulatory approval. Terms not disclosed; one outlet reports $1.2B, unconfirmed. Source: Nordic Capital, 7 Oct 2026. |
| ConductorOne (C1.ai) | Acquired by Workday | Undisclosed | 13 Oct 2026 | Agentic identity platform that governs access for people and AI agents across enterprise systems. Definitive agreement dated 12 October, announced at Workday Rising on 13 October; expected to close in Workday's fiscal Q4 2027 (ending 31 January 2027). Terms not disclosed. Source: Workday (SEC Form 8-K, Exhibit 99.1), 13 Oct 2026. |
| Anon | Acquired by Sierra | Undisclosed | 5 Nov 2026 | Agent authentication and browser infrastructure. Terms not disclosed; Anon says it raised more than $13M. Source: Sierra, 5 Nov 2026. |
| Level AI | Acquired by Five9 | $325.0 million | 5 Nov 2026 | AI-native customer experience platform. Definitive agreement, about $325 million in cash from Five9's balance sheet; expected to close in 2026, subject to closing conditions, so it is agreed, not completed. Source: Five9 (SEC 8-K exhibit), 5 Nov 2026. |
Amounts count only when the buyer or the company disclosed them; reported prices stay in the detail column and never enter a total. Data through 13 November 2026.
Two deals from the same window are left out on purpose. HCLSoftware agreed to buy Robotiq.ai, a robotic process automation platform, and LiveKit bought Loophole Labs, a virtualization company; both buyers bought them for agent work, but neither target sells an agent product. ServiceNow's purchase of Sweep and Palo Alto Networks' purchase of Console were both announced in the first days of September, outside this window, and belong to the Q3 story.
Three rounds that close out Q3
The Q3 tracker's 28 September pass stopped its September section at 24 September. Three rounds landed after that and belong to Q3. This edition records them, and the Q3 page was closed and restamped in the same batch on 4 October 2026, so both pages agree.
| Company | Amount | Stage | Valuation | Date | Detail |
|---|---|---|---|---|---|
| Instinct | $1,000M | Series C | $10B | 28 Sep 2026 | Consumer AI agent. TechCrunch reports the company confirmed the round type in its own press release. The invite-only service launched in August 2026, which makes this a very fast growth round. |
| Reco | $55M | Growth | Not disclosed | 29 Sep 2026 | Agent security: finding and securing the agents already running inside a company. Builds on a $30M Series B in February 2026, taking total raised to $140M. |
| Restate | $20M | Series A (additional) | Not disclosed | 30 Sep 2026 | Durable execution runtime for long-running agent workflows. Reported as led by Singular. I treat that lead as reported, not confirmed. |
Three rounds, $1,075.0 million, all dated 28 to 30 September 2026 and all counted in Q3.
On Instinct's investors: TechCrunch's piece is headlined as a $1 billion Series C at a $10 billion valuation and reports the company confirming the round in a press release. A Bloomberg article from the same day reached me only as a search summary, so no investor names go in the table. The amount, the stage and the valuation are what I can stand behind.
Reco's valuation is a good example of why this page keeps valuations in a separate column. CEO Klein said the valuation has "more than doubled" since February and placed it vaguely in the "high hundreds of millions" without giving a figure. That is useful context and it is not a number. It stays in prose and it never enters a total.
Restate is the smallest of the three and the one I find most interesting, because it is the second durable-execution round in the Q3 window. Temporal Technologies raised $550 million on 14 September for the same layer of the stack. Two rounds for the plumbing that keeps a long-running agent from losing its place is a signal about where the hard engineering problems actually are.
Related reading: the AI agent adoption statistics for 2026 page tracks the demand side with the same sourcing rules, and the AI agent pricing index tracks what these companies charge once the money is in.
Q3 2026 final close
Q3 2026 closes in this compilation at 59 disclosed rounds and about $6.32 billion. These totals are my own compilation of publicly disclosed rounds. They are not a market report, they are not reconciled against Crunchbase or PitchBook, and they do not claim completeness. The Q3 tracker carries the per-round source for every line that makes up the figure.
The arithmetic, shown
Here is the whole calculation, because a total nobody can reproduce is a total nobody should cite.
| Month | Rounds at 28 Sep pass | Added here | Final rounds | Final amount |
|---|---|---|---|---|
| July 2026 | 16 | 0 | 16 | $851.0M |
| August 2026 | 17 | 0 | 17 | $1,185.8M |
| September 2026 | 23 | 3 | 26 | $4,284.2M |
| Q3 2026 total | 56 | 3 | 59 | $6,321.0M |
September: $3,209.2M at the 28 September pass, plus Instinct $1,000M, Reco $55M and Restate $20M, giving $4,284.2M. Q3 moves from $5,246.0M to $6,321.0M, about $6.32 billion.
One note on rounding, since this page gets quoted. Rounding the 28 September subtotal to $5.25 billion first and then adding $1.075 billion yields $6.325 billion. The month-level sum is $6,321.0 million. I treat the month-level sum as authoritative and quote it as about $6.32 billion.
Cognition excluded: 58 rounds and about $4.32 billion. Cognition closed on 8 September at what the company calls over $2 billion, led by Andreessen Horowitz and Accel, at a $48 billion valuation. No exact figure was published, so it is recorded at a $2 billion floor. That single round is roughly a third of the quarter, which makes the unadjusted $6.32 billion a poor input for any quarter-over-quarter comparison. Use $4.32 billion across 58 rounds when you compare Q3 to another quarter, and say which basis you used.
The four largest Q3 rounds
Each of these is already sourced on the Q3 page, line by line. Three of the four sell to developers or enterprises: coding agents in two cases and the execution layer underneath them in the third. The fourth is the consumer exception.
| Company | Amount | Valuation | Date | Lead | What it sells |
|---|---|---|---|---|---|
| Cognition | Over $2,000M | $48B | 8 Sep 2026 | Andreessen Horowitz, Accel | Coding agents. Recorded at a $2,000M floor because no exact figure was published. |
| Instinct | $1,000M | $10B | 28 Sep 2026 | Not recorded | A consumer AI agent that completes tasks. Added by this edition. |
| Temporal | $550M | $12.55B | 14 Sep 2026 | Lightspeed (co-lead) | Durable execution: the layer that keeps long-running agents alive across failures. |
| Factory | $200M | $5B | 15 Sep 2026 | Not recorded | Coding agents. The round the Q3 page missed twice before catching it. |
The four rounds above are $3,750.0 million of the $6,321.0 million Q3 total, roughly three-fifths of the money across four of the 59 rounds.
AI agent funding history, as far as this page has compiled it
This franchise has covered the period since Q2 2026, and compiled it round by round since Q3 2026. The table below covers both, and that split is the honest answer to the question people actually arrive with.
There is no 2022, 2023, 2024 or 2025 total on this page. Those years are absent because this page publishes only figures it has compiled itself against named sources, and I have not compiled them. Several third-party totals for those years exist and I could paste one in, which would make this section look complete and would make every figure on it unverifiable by the standard applied to the rest of the page. A number you cannot rebuild from a table on this site does not get printed on this site.
| Period | Rounds compiled | Disclosed total | Basis |
|---|---|---|---|
| Q2 2026 | Not compiled round by round | No total published | Qualitative recap only, folded into the Q3 tracker. Headline round: Cognition, over $1,000M at a $26B valuation, late May 2026. |
| Q3 2026 | 59 | $6,321.0M | Full month-by-month compilation, final. July $851.0M, August $1,185.8M, September $4,284.2M. |
| Q3 2026, Cognition excluded | 58 | $4,321.0M | The comparable basis, from the same arithmetic. Removes the $2,000M floor recorded for a round with no published exact figure. |
| Q4 2026 to date | 18 | $1,656.0M | Open quarter, from the Q4 rounds table. Data through 13 November 2026, so treat it as a floor. |
Every row is a Gravity compilation of publicly disclosed rounds. Each figure links back to the table it was built from. Jump to any row with its fragment, for example #stat-q3-2026.
Why Q2 2026 has no round count
The Q2 2026 edition of this tracker was written as a four-fact recap of the quarter. It names Cognition's late-May raise of over $1 billion at a $26 billion valuation, it records the valuation jump from $10.2 billion in roughly eight months against a $492 million revenue run-rate, and it describes where the mega-rounds went. It stops short of listing every disclosed round, so there is no count to carry forward.
I am leaving that gap visible. Back-filling Q2 now would mean compiling a quarter five months after it closed, from coverage that has already been reorganised by whatever happened since, and the result would be weaker than what the Q3 and Q4 tables can claim. The Q3 tracker holds the Q2 recap in full if you need the qualitative picture.
What changes across the compiled quarters
Two quarters of comparable data is thin, and I will say what it supports and stop there. Q3 2026 produced a $6.32 billion compiled total across 59 rounds, with the top four rounds accounting for roughly 59 percent of the money. That is concentration at the top of the table.
Q4 2026's first two weeks, to the 16 October 2026 cut-off, continued that shape: two rounds of $500 million or more (Manus, Armadin), a $52.5 million and a $60 million round, and a $4.5 million seed. Whether the shape holds is the open question, and it is the thing the next few monthly updates will actually answer. For the demand side of the same period, the AI agent adoption statistics page tracks deployment with the same sourcing rules, and the AI agent pricing statistics page covers what buyers in that market actually pay.
What the pattern says
Three observations hold up against the rounds listed above. I have deliberately stopped at three, because a fourth would need a number I do not have.
Agent security is becoming its own funding category
Reco raised $55 million on 29 September and Armadin raised $255.5 million on 1 October. Back to back, straddling the quarter boundary, $310.5 million between them. TechCrunch's own framing on the Reco piece was that AI agent security startups are crowding the market, which is a notable thing for a technology publication to say about a category rather than a company.
Look at what those two companies do. Reco finds and secures the agents already running inside a company. Armadin chains security weaknesses into attack paths so a defender can test them. Both sell the same assumption: agents are already deployed, already acting, and already a surface worth attacking. That assumption is now worth nine-figure cheques.
The runtime layer keeps raising
Temporal took $550 million on 14 September and Restate took an additional $20 million on 30 September. Both sell durable execution: the infrastructure that lets a process run for hours or days, survive a crash, and resume where it stopped rather than at the beginning.
Durable execution is the right problem to be funding, unglamorous as it sounds. An agent that researches a market for six hours and then loses its place is worse than useless, because you cannot tell which half of the work is real. I have watched this failure mode from the build side and it is why the money is going to the plumbing. If you want the conceptual version of what that layer does, the AI agent architecture patterns breakdown walks through where runtime sits relative to the agent itself.
Consumer agents got a billion-dollar data point
Instinct launched an invite-only service in August 2026 and raised $1.0 billion at a $10 billion valuation on 28 September. Weeks, not years. TechCrunch framed the company as part of a new class of consumer agents that get things done instead of answering questions.
One round does not make a category, and that is the honest reading. What it does is set a reference price. Until 28 September there was no billion-dollar consumer-agent data point, and the absence of one was a reasonable argument that the money only believed in enterprise and infrastructure. That argument now needs a different piece of evidence. If you are thinking about what a consumer agent replaces in practice, the AI employee framing is the closest adjacent read.
What to watch in Q4
No numbers attached to any of these and no forecast. The current pace sits in the headline numbers; anything after the data cut-off stated there is unchecked.
- Whether agent-security funding keeps the cadence it set across the Q3 to Q4 boundary, or whether Reco and Armadin turn out to be two rounds that happened to land three days apart.
- A second consumer agent raising at a comparable scale after Instinct, which is the single cleanest test of whether that round was a category signal.
- Whether the $30 million median cheque from the Q3 compilation holds as Q4 fills in.
- Repeat raises from the durable-execution companies, and the stage they come at.
- Whether a round that gets reported as closed in Q4 turns out to be talks, which happened to Cognition for four weeks on the Q3 page.
Reading funding tables is itself a task worth handing off. Explore agents and describe the research you want run. Free to start, no card, paid plans from $15 a month on pricing.
Methodology
I compile this tracker and I check every round against a named source. I did not speak to any founder or investor for this edition, and nothing here comes from a private conversation. Where a figure is weaker than it looks, the page says so in the same sentence as the figure.
What counts as a round
A round enters a table when three things are true. The company is named, the amount is disclosed, and either a named publication reported it or the company published it itself. A company announcement, a lead investor's own post, or coverage by a major outlet all clear that bar.
An aggregator on its own falls short of that bar. A funding newsletter or a directory's daily brief is how I find a round, and the citation goes to the publication that brief credits. Where I reached a round only through an aggregator, both the brief and the publication it credits are named in Sources so you can see the full chain and judge it yourself. Three of the rounds on this page arrived that way and the chain is printed for all three.
Several things fail that test and get excluded on purpose. Reported talks are not a closed round, and the Q3 page held Cognition's reported raise out of its totals for four weeks on exactly that basis. Debt facilities are not equity rounds. A company whose product is adjacent to agents but is not an agent product gets left out, which is why a $150 million Postgres platform round announced in the same window as the Q4 rounds above does not appear here. Rumours with no primary source do not appear at all, however widely they circulate.
Valuations, ranges and floors
A valuation lives in its own column and is never added to a total. The most common error in funding coverage is a valuation repeated as a raise, usually because the headline number is bigger and reads better. When I find that error in a source, the page notes it rather than quietly dropping the round.
Where a company gives a range or an approximation, I record the floor and label it. Cognition is the clearest case on either tracker: the company says "over $2 billion" and no exact figure exists, so it sits at $2,000 million and that is the one number in the Q3 total that is deliberately too low. A floor you can see is auditable, so a floor is what the page prints.
Where a lead investor is reported but not confirmed, the table says reported. Where I could not verify an investor at all, the column stays empty. Restate's lead is in the first category and both Q4 rounds are in the second.
How the headline numbers are computed
The headline numbers are recomputed every day by a script that reads the round tables on this page and on the Q3 tracker. Nothing in that block is typed by hand. The as-of date is the latest data cut-off stated on the page, so a day with no new research does not pretend to be fresher than it is.
- Totals add disclosed amounts only. A round with an undisclosed amount counts as a round and adds nothing to the money.
- Pace is rounds divided by days elapsed in the quarter up to the cut-off. The same-day comparison counts Q3 rounds dated on or before the matching day of Q3, so day 4 of Q4 is set against 1 to 4 July.
- Median and average appear once the quarter has five disclosed rounds, because below that they describe no round that happened.
- Category share uses the Category column of the Q4 table (Security, Agent infrastructure, Coding agents, Consumer agents, Vertical agents, Agent platforms) and appears with the median.
- Exits count an acquisition when the buyer or the company announced it, including signed agreements that have not closed. A reported price stays in the detail column and never enters a total.
The full Q4 table is in data.csv, one row per round, with amounts in US dollar millions and a flag for amounts recorded as a floor.
How a correction gets logged
Corrections go in the update log with a date, and the figure that changed stays visible in that line. A reference page earns the name by keeping its own history readable.
The Q3 tracker is the proof that this matters. It was revised upward five consecutive times, always in the same direction, and 15 September 2026 alone was corrected three separate times. The third correction found Factory's $200 million round, which two earlier passes had missed on a day they had both already reviewed. The Q3 close in this edition is the sixth upward revision of that quarter. Treat any open-quarter total here as a floor, including the Q4 figure at the top.
If a round is missing, send it through contact with a dated source and it goes in on the next pass, with a line in the log saying where it came from.
Why Gravity has not raised
Gravity has raised nothing. It is bootstrapped and built in Bangalore, and I run it as the fourth company I have started after shutting three down.
That history is the reason. Each previous time, outside expectations pulled the product toward what the next deck needed and away from what the people using it needed. Bootstrapping leaves one constituency to answer to. It also means I read a funding announcement as a list of new obligations a competitor has taken on, which is probably why this page spends more words on what a round proves than on how big it was.
The product is in public beta. If you want to see what an agent run looks like before you start, walk through a run first. For how the costs compare across the category, the cheapest AI agent platforms breakdown is dated and maintained the same way this tracker is.
FAQ
How much AI agent funding has been disclosed in Q4 2026 so far?
18 disclosed rounds worth $1,656.0 million, with data through 13 November 2026. The largest is Manus's $500.0 million Not stated round on 8 October 2026. The quarter is open, so that figure is a floor and it moves on every weekly pass.
What was the final Q3 2026 AI agent funding total?
In this compilation, 59 disclosed rounds and about $6.32 billion: July at 16 rounds and $851.0 million, August at 17 rounds and $1,185.8 million, September at 26 rounds and $4,284.2 million. Exclude Cognition, recorded at a $2 billion floor, and the quarter is 58 rounds and about $4.32 billion. The Q3 tracker holds the source for every round in that sum.
Which AI agent startups raised money in November 2026?
Through 13 November 2026, eight: Finpilot, a $10.0 million Series A in vertical agents; firmly, a $15.0 million Series A in vertical agents; Bay, a $37.0 million Series A in security; Daylight Security, a $80.0 million Series B in security; Lumbra, a $25.0 million Series A in agent platforms; Plurai, a $13.0 million Seed in agent infrastructure; Rox, a $100.0 million Series B in vertical agents; and LangChain, a $300.0 million Series C in agent platforms. The rounds are dated 4 November 2026 to 13 November 2026.
Which AI agent startups were acquired in late 2026?
As of 13 November 2026, seven: Salesforce agreed to acquire Listen Labs on 29 September, ZoomInfo acquired DoubleO.ai on 30 September, Inworld acquired Ultravox on 1 October, Automation Anywhere agreed to acquire Boost.ai on 7 October, Workday agreed to acquire ConductorOne (C1.ai) on 13 October and Five9 agreed to acquire Level AI on 5 November, and 1 more. One of these deal disclosed a price. The acquisitions and shutdowns section keeps the running list with a source for each deal.
Why did the Q3 2026 total change after the quarter ended?
The Q3 tracker's 28 September pass stopped its September section at 24 September. Three rounds landed on 28, 29 and 30 September: Instinct at $1.0 billion, Reco at $55 million and Restate at $20 million. Adding them moved Q3 from 56 rounds and about $5.25 billion to 59 rounds and about $6.32 billion.
Does this tracker count a valuation as funding?
No. A valuation sits in its own column and never enters a total. The most common error in funding coverage is a valuation repeated as a raise, so a figure only counts when the publication or the company describes it as money raised.
How often is this page updated?
Weekly, on the Monday pass, for as long as Q4 2026 is open. Every total carries the date its research window closed, and corrections are added to the update log rather than edited in silently.
Is Gravity funded?
No. Gravity is bootstrapped, built in Bangalore, and has raised no outside capital. The product is in public beta, it is free to start with no card, and paid plans start at $15 a month on the pricing page.
Can I reuse the numbers on this page?
Yes. The round data in the tables is published under CC BY 4.0, so reuse it with a link back to this page. If a round is missing, send it through the contact page with a dated source and it goes in on the next pass.
Update log
- Weekly pass, three rounds and one exit. Searched 5 to 16 October (the 12 October pass did not reach the data files). Added Stuut ($52.5M Series B, 7 Oct), Manus ($500M floor, 8 Oct; the $4B valuation is reported, not confirmed) and Hone ($60M seed, 8 Oct). Logged Workday's agreement to acquire ConductorOne (C1.ai). Not added: Zenzap's $33M Series A (the report's own timing is ambiguous), Melius and Catalyst (outside the agent scope), Nous Research and doxx.net (no primary page read).
- Weekly pass, no new rounds. Searched 27 September to 5 October. No new Q4 AI agent round with a primary source beyond Armadin and Photon. Armadin's leads (Andreessen Horowitz, Accel) and its $2.5 billion-plus valuation are now recorded from TechCrunch. Not added: EliseAI's reported $350 million round of 29 September (no primary announcement reached, borderline scope, Q3), and General Intuition (game-trained models, not an agent product).
- Published as a live tracker. Added the headline numbers block (recomputed daily from the tables), a Category column on the Q4 table, a downloadable data.csv, and the acquisitions and shutdowns section, seeded with Salesforce and Listen Labs (29 September), ZoomInfo and DoubleO.ai (30 September) and Inworld and Ultravox (1 October).
- First edition. Q4 opened with 2 rounds and $260.0 million, data through 5 October 2026. Added Instinct, Reco and Restate to the Q3 close, which moves Q3 to 59 rounds and about $6.32 billion. Armadin's reported co-leads and a reported $80 million round from a separate company were both excluded for lack of a primary source, and a $150 million Postgres platform round was excluded as out of scope.
Every future pass appends a line here with its own date and its own window. If a number on this page changes, the reason it changed will be in this list.
Two sections are deliberately unbuilt at this edition. November 2026 and December 2026 are labelled empty and get their tables at their monthly updates. The funding history section starts at Q2 2026 because that is where this franchise's coverage starts, with the round-by-round compilation beginning a quarter later in Q3 2026.
Previous editions
For Indian agent startups specifically, round by round with a stated method, see the India AI agent funding tracker.
- AI agent funding tracker, Q3 2026: closed at 59 disclosed rounds and about $6.32 billion, with a source on every round. The Q2 2026 edition was folded into it and its recap lives there.
Sources
- TechCrunch: Viral AI agent Instinct raises $1B Series C at a $10B valuation (September 28, 2026)
- TechCrunch: Reco raises $55M as AI agent security startups crowd the market (September 29, 2026)
- TechCrunch: Kevin Mandia's new 'agent swarm' security startup Armadin raises $255.5M at $2.5B valuation (October 1, 2026)
- aiagentsdirectory.com: AI Agents News Brief, October 1, 2026, carrying Tech Funding News on Photon (1 October 2026) and AI Weekly on Restate (30 September 2026).
- aiagentsdirectory.com: AI Agents News Brief, October 2, 2026, carrying Tech Funding News (2 October 2026) and Superpower Daily (1 October 2026) on Armadin.
- The Next Web: Stuut raises $52.5M Series B to run order-to-cash with AI (October 7, 2026)
- SiliconANGLE: AI agent developer Manus raises $500M+ at reported $4B valuation (October 8, 2026)
- Bloomberg: Former OpenAI, Cognition staffers want AI to help run a business (October 8, 2026), read via Runtimewire, on Hone.
- Workday, SEC Form 8-K Exhibit 99.1: agreement to acquire C1.ai / ConductorOne (October 13, 2026)
- GeekWire: Finpilot raises $10M for AI agents that handle the grunt work for institutional investors (November 4, 2026)
- Retail Technology Innovation Hub: firmly announces $15 million Series A led by Squarepoint Capital (November 6, 2026)
- TokenPost: Bay raises $50 million to expand its AI agent security platform (November 11, 2026; Bay announced November 10)
- ultrathink.ai: Rox raises a $100M Series B at a $1.3B valuation (November 11, 2026), citing the company's PR Newswire release, Fortune and Reuters
- SiliconANGLE: Daylight raises $80M on 40-fold revenue growth (November 10, 2026)
- Business Wire, company release: Lumbra raises $25M Series A (November 10, 2026)
- PR Newswire, company release: Plurai raises $13M (November 11, 2026)
- LangChain, company blog: Our Series C announcement, $300M at a $3B valuation (reported by TechCrunch on November 13, 2026)
- Sierra: Sierra acquires Anon (November 5, 2026)
- Five9, press release filed as SEC 8-K exhibit: agreement to acquire Level AI (November 5, 2026)
- Gravity, AI agent funding tracker Q3 2026, updated 4 October 2026 (Q3 close). Own compilation; per-round sources for July, August and September are listed there, including Cognition, Temporal Technologies and Factory.
Excluded from the Q4 table and named here for the record: a reported $80 million Series A that no primary source confirmed, and a $150 million round for a Postgres platform company, which is infrastructure but not an agent product.