Agentic AI companies in 2026: 20 checked against five criteria, with 3 meeting all five
Search for agentic AI companies and most of the lists you'll find were written by agencies that sell agent builds: five of the six on Google's first pages put their own firm first, and the sixth puts itself first in the group it recommends.
We took every firm those lists, Clutch's directory and Google's AI overview name, and checked the 20 that at least two of them name or that Google shows with a service page of their own. Criterion 3 narrowed the field most, and it's the plainest: show, on your own site, an agent you built for a client you name, with a result. Three firms met all five criteria:
- HyperSense Software, for one agent shipped in about a month: a billing agent for German landlords that, it expects, turns 20+ hours of work into 15 minutes.
- Azumo, for requests that arrive by email: Angle Health's quotes, down from 45 minutes each to 5.
- EffectiveSoft, for agents inside an engineering process you already run: Motive Retail's integration agents.
Meronym, which publishes this list, builds AI agents for clients, so every firm here is a competitor, and we have no referral or partner deal with any of them. Our own entry is fourth, marked as ours, with the same fields as the others. It isn't one of the three: today it meets the first two criteria and not the third. We read every firm's own pages and Clutch's directory on 27 September 2026, using the Internet Archive's latest copies for the three sites that blocked our browser; we didn't call any firm, run a trial or check a reference.
Who this list is for
It's for a business hiring a firm to build an AI agent that works in its own systems. Three other readers searching the same words should go elsewhere:
- Buying a ready-made agent? Our AI agents for business guide compares the products by job, and our enterprise AI agents guide compares the platforms by what you already run.
- Running a multi-year programme with a global integrator? Accenture, IBM Consulting, Cognizant, Infosys and TCS appear across three of the lists, but none shows a minimum project on Clutch or a starting price on its home page, so you'll compare their bids, not their pages. They're included below among the firms that did not meet every publication criterion.
- Looking for stocks? Google's "People also ask" box on this search asks for the best agentic AI companies to invest in. Every firm here sells services, so this list won't help.
What the ranking lists and Clutch don't tell you
The lists rank their own publishers. SoluLab's puts SoluLab first, DevCom's DevCom and Intellectyx's Intellectyx. Inoxoft's is a video that sends you to a longer guide, and both put Inoxoft first. Lindy's, which now covers tools rather than firms, puts Lindy first, and The JADA Squad's says "This is not a ranked list" and puts itself first in the group it recommends to mid-market and enterprise teams. None says which of the firms it lists it has worked with.
These are the lists on Google's first page for the two searches as Ahrefs last recorded them in the US: on 5 September 2026 for "agentic ai companies" and on 29 July 2026 for "ai agent development companies". Every Google position on this page comes from those two records, so the page you see may differ.
Clutch's directory, second on Google for "ai agent development companies", sorts its AI agent developers by "Sponsored" unless you change it. Its own note on that sort: "Sponsored providers pay us a fee for placement within our listings." On 27 September the first nine firms on its first page were sponsor listings. A tenth, labelled "Featured", links through Clutch's pay-per-click system, and the services that firm lists on Clutch include no AI agent work.
Clutch's ratings don't separate the firms either: all 51 on that first page score between 4.7 and 5.0, and 49 score 4.8 or higher. What does separate them is the share of each firm's listed services that are AI agents. We counted it for all 51: 33 put agents at 15% or less of the services they list, and only 7 at 30% or more.
Gartner estimated in June 2025 that only about 130 of the thousands of agentic AI vendors are real, and blamed "agent washing": assistants, RPA tools and chatbots rebranded as agents. A place on a list, or at the top of a directory page, tells you little about which kind you're talking to. What a firm can show you does, and that's what the rules below test.
How we chose: five published-evidence criteria
We started from every name on the six ranking lists above (Inoxoft's video and its guide count as one), on Clutch's first page, in Google's AI overview for "ai agent development companies" and on the three agency pages ranking for it: 99 names. Seventeen are products or platforms, Lindy, CrewAI and Microsoft Copilot Studio among them, which left 82 firms that build agents for clients. We checked 20: the 18 that at least two of those sources name, not counting the firm's own list or page, and two more that Google shows with a service page of their own (Azumo's in the top ten, Appinventiv's in the AI overview). Three met all five criteria:
- It sells agent builds to clients. A page offering to build agents for you, not a product you pay a licence for.
- It publishes a price floor. A minimum project size on its Clutch profile, or a starting price on its own site, so you know whether you can afford it before a call. An hourly rate alone doesn't give you the floor.
- It shows an agent it built, for a client it names, with a result. On its own site, all three: a client you can look up by name, not a description such as "a US bank" or a label with no company behind it; an agent, meaning a language model does part of the job and the software then acts on it, such as preparing a statement, creating a quote, writing code or making a transfer, rather than a chatbot that only answers or a script with no model; and a figure for the work it took off people, in time, money, volume or errors. A forecast or a client's own estimate counts if the page labels it, and we label it too. Engagement, app speed and usage figures don't count, and neither does an agent the firm built for itself.
- It says what waits for a person. The case study for the agent it shows names something the agent can't finish without a person, such as an approval, a confirmation or a hand-off to staff. A promise about the firm's usual practice, made elsewhere on its site, doesn't count.
- Someone else vouches for it. Verified reviews on Clutch, or a place on two lists it didn't write.
We applied the criteria in order, so each firm that did not meet all five is listed under the first unmet criterion. Criterion 2 narrowed out the five global integrators, criterion 3 eleven firms and criterion 4 SoluLab. No firm stopped at criterion 1 or 5. Criterion 4 asks only that the case name one such step, not whether the step is enough, so each entry quotes it and you can judge how far it goes. A firm may have strong unpublished work and still not meet criterion 3; the questions further down help you ask for that evidence directly.
How we chose: 99 names, 20 checked, 3 met every criterion
Each entry gives the Clutch review count with the rating in brackets. All three that passed score 4.9, so the ratings decide nothing here. Clutch's headcounts don't agree with themselves: HyperSense's card says 10 to 49 people, while the office list on the same profile adds up to 86 to 115. So each entry says where the firm's offices are, not how many people it has. Our own entry passes rules 1 and 2: it publishes a price floor, but with no client agent published it can't meet rules 3 and 4, and it has no reviews yet. The list isn't ranked: it's grouped by the kind of job each firm's evidence fits, and ours by how we scope and price a build, with our entry fourth.
One agent, shipped in about a month
HyperSense Software: best for one well-scoped agent, built and live in weeks
The agent it shows you. Reperks, a German property-tech startup, whose agent, Perks, prepares the annual utility-cost statements German landlords owe each tenant. HyperSense expects a landlord to finish in 15 minutes instead of 20+ hours: the case study files that figure under "Business impact and expected results", with no count of landlords behind it, and its agent page gives 10 to 15 minutes per billing cycle. Reperks' own site claims less for its product: "80% weniger manueller Aufwand pro Abrechnung", 80% less manual work per statement. The production system took a six-week sprint, and it's live at agent.reperks.de.
What waits for a person. In the case study's words, "Extractions below 98% confidence trigger user confirmation flows", so the landlord confirms anything the agent wasn't sure it read, and "You see and approve everything before sending to tenants". Reperks' own site describes the same step: the landlord checks the exceptions the agent flags and releases the ones that need a person's decision. The calculations run in ordinary code, checked three independent ways, not in the model. On its agent page, HyperSense says it starts each build by asking "What must it never do? Where do humans approve?", and that another agent it built, ZURI, runs "5,000+ automated assertions on every code change".
Price floor. A $25,000+ minimum project and $50 to $99 an hour on Clutch. Its AI Feature Sprint is "one month to define, build, and deploy a single high-impact agent", with no price on the page.
Team and focus. Its Clutch profile marks Schaumburg, Illinois, as headquarters and lists a second office in Bucharest, but the firm's own overview on that profile describes "a 40+ person team headquartered in Bucharest with a US office in Chicago". AI agents are 40% of the services it lists on Clutch, the largest share of the three.
Who else vouches. 31 Clutch reviews (rated 4.9); also on Inoxoft's list.
- Its showcase agent is live at a public address you can open
- Agents are a bigger share of its listed services than at any other firm here
- The 15-minute figure is an expected result, with no count of landlords behind it, and the client's own site claims less
- Its showcase is a startup's own product, not an agent inside an established company's systems
Paperwork and intake: documents in, records out
Azumo: best for reading requests that arrive by email and drafting the quote
The agent it shows you. Angle Health, a US health-insurance technology company whose requests for proposals arrive by email as Zendesk tickets. Azumo's workflow sorts each request's attachments by what they contain and sends the relevant information to a language model (currently GPT) to pull out the details, then creates the quote. Azumo reports that the time per request fell from 45 minutes to 5, and that the workflow "is now being evaluated for patenting".
What waits for a person. In the case study's words, "Users validate the generated quote before underwriting approval", and sales sends the proposal only once underwriting approves it. On its agent page, Azumo says its agents run "fully autonomous for low-risk tasks, human-in-the-loop above defined thresholds", and that every deployment includes "fallback paths (what happens when it fails)".
Price floor. A $10,000+ minimum project and $25 to $49 an hour on Clutch: the lowest minimum and the lowest rate band of the three.
Team and focus. Headquartered in San Francisco, with a second office in Rosario, Argentina, by its Clutch profile. AI agents are 15% of the services it lists on Clutch.
Who else vouches. 27 Clutch reviews (rated 4.9); its agent page ranks sixth on Google for "ai agent development companies".
- A person approves every quote the system drafts
- The lowest minimum and the lowest hourly band of the three
- The case study calls it LLM automation: a model reads the request, and fixed steps build the quote
- It reports time per quote, with no volume or error rate
Agents inside an engineering process you already run
EffectiveSoft: best for adding agents to an engineering process with its own review standards
The agent it shows you. Motive Retail, which already used EffectiveSoft as its engineering partner, wanted new automotive integrations onboarded faster without changing its engineering and validation standards. EffectiveSoft's Claude agents now do the scoped development work for each integration from a specification that senior engineers have reviewed and approved. The agents run as a plugin called Lynx, and EffectiveSoft reports that a month after rollout it made about 205 tool calls a week, and that Motive Retail "models the time saved per assisted conversation at 4 hours".
What waits for a person. Senior engineers approve each specification before the agents work from it, and review what the agents produce: "AI-generated specifications and implementation drafts must pass engineer review before moving downstream." EffectiveSoft's agent page offers "human-in-the-loop (HITL) control" and says it monitors deployed agents "to detect performance issues, prevent drift".
Price floor. A $25,000+ minimum project and $50 to $99 an hour on Clutch.
Team and focus. Headquartered in San Diego, with offices in Warsaw, Pittsburgh and San José, Costa Rica, by its Clutch profile. AI agents are 20% of the services it lists on Clutch.
Who else vouches. 19 Clutch reviews (rated 4.9), the fewest of the three; on Inoxoft's list. Its place on Clutch's first page is a sponsored listing, the only one of the three.
- Engineers approve the specification before the agents start, and review their drafts after
- In regular use: about 205 tool calls a week, a month after rollout
- The time saved is the client's model, not a measurement
- The agent serves an engineering process, so it shows less about agents in customer-facing or back-office work
Written limits and a fixed quote before the build
Meronym (the publisher of this list): best for buyers who need a fixed price and the approval rules in writing before any build starts
The agent it shows you. None published yet. That is why our own entry does not pass the evidence rule used for this list.
What waits for a person. We agree in writing what the agent may do alone, what needs a person and what it must refuse, and write actions wait behind an approval gate. On our AI agent development page, production agents start at the level where the agent "stops at a gate for anything above a threshold", and move up only when the scores hold on a test set built from your own past cases, rerun on every change.
Price floor. $8k: a pilot on one workflow is $8k–$35k and a production agent $16k–$82k, on our AI agent development services page. The quote is fixed at the end of a two-week scope, $4k–$12k and credited against the build, with build and running costs shown separately and the engineer who will build it named.
Team and focus. AI agents, generative AI and automation, built and run by our own engineers. This page does not publish a team size or office breakdown.
Who else vouches. This page cites no independent review profile or review count. We're not on Clutch's first page or on any of the lists above.
- The approval rules and the test set are agreed before the quote
- You own the code, prompts and test sets at handover
- No published client agent, price or independent reviews yet: this entry passes rule 1 only
The four side by side
Screenshot this one: each firm's proof and price floor in one place.
| Firm | Best for | The agent it shows | Result it reports | Minimum project | Hourly rate | Agents' share of its services | Clutch reviews (rating) |
|---|---|---|---|---|---|---|---|
| HyperSense Software | One agent in about a month | Reperks: landlords' utility statements | 20+ hours to 15 minutes a landlord's annual billing, expected | $25,000+ | $50 to $99 | 40% | 31 (4.9) |
| Azumo | Email requests into quotes | Angle Health: RFP to quote | 45 to 5 minutes a quote | $10,000+ | $25 to $49 | 15% | 27 (4.9) |
| EffectiveSoft | Agents inside an engineering process | Motive Retail: integration onboarding | 4 hours saved per assisted conversation, modelled by the client | $25,000+ | $50 to $99 | 20% | 19 (4.9) |
| Meronym (publisher) | A fixed price and written approval rules first | None published yet | No published client result yet | $8k, a pilot on one workflow | Not published | Not published | No independent review count cited |
Source: each firm's own case study, linked in its entry, and Clutch's AI agent directory, first page. Checked 27 September 2026.
The seventeen that did not meet every publication criterion
Sixteen of these are on the ranking lists, and Google's AI overview cites the seventeenth. Their placement reflects only the evidence published on their sites, not what they may be able to build; ask any of them the questions in the next section.
No published price floor (rule 2)
- Accenture, IBM Consulting, Cognizant, Infosys and TCS are named on DevCom's, Intellectyx's or The JADA Squad's lists, but none shows a price floor. Cognizant's Clutch profile gives $200 to $300 an hour with the minimum "Undisclosed", TCS's shows both as "Undisclosed", we found no Clutch profile of their own for Accenture, IBM or Infosys, and none of the five home pages shows a starting price (Infosys's and TCS's blocked our browser, so we read the Internet Archive's September copies). IBM does publish prices for its watsonx Orchestrate platform, which our enterprise AI agents guide covers, but that's a product you pay a licence for, not a build.
- What four of them show at rule 3. IBM names a client for its agent platform: Inter, a Brazilian bank that moved its 18 assistants onto watsonx Orchestrate, with "an estimated 95% reduction in deployment effort". IBM credits that figure to automating the bank's deployments through the platform, so it measures the platform, not the agents' work. Accenture's Best Buy assistant can change a customer's delivery or subscriptions, with no figure. Infosys's Financial Times case, read in the Internet Archive's copy of 9 June 2026, reports 65 hours a month of service staff's time saved in its pilot phase by automatic case classification and suggested replies, and presents agentic AI as the next step. Cognizant's agent results are for clients it describes rather than names, such as "a UK financial services firm". We didn't read further for TCS.
No agent for a named client with a result (rule 3)
- Intuz: its case study has specific figures, order processing cut from five minutes to 30 seconds, but describes the client only as "A home health provider". "careonix" is a label on the page, not a home health provider we could find, and the Clutch review it quotes is anonymous.
- Qubika, Clutch's first sponsored listing: its agent examples are its own tools (QBricks "provides both pre-built agents as well as a workflow builder"), and its lender case study names no client and covers data and machine-learning work.
- LeewayHertz, cited in Google's AI overview and third for "ai agent development companies": its case studies are LLM apps, one of them for Scrut, with no agent carrying a client's name and a figure, and its agents are sold as "ZBrain AI Agents". The copyright line on its site is The Hackett Group's.
- Appinventiv, also cited in the AI overview: Tootle's figures are "a 35% increase in user engagement" and "10x faster responses", which measure use and app speed rather than work taken off anyone, and its MyExec case gives no figure.
- Trigma, a Clutch sponsor: its voice-agent case was built for an unnamed "US-based healthcare revenue cycle management (RCM) company", whose testimonial on the page says "we've just wrapped phase one", and its other agent cases describe Trigma's own platforms.
- Markovate: its ERP agent reports 95% accuracy for a client it doesn't name.
- Quytech: its agent cases give figures for clients described, not named, as an "Indian Consumer Product Company" (material planning) and a "UK-Based Startup" (voice ordering).
- Software Mind: its agent case describes the client only as "One of the world's leading health and security risk services company".
- Osedea: Genius Solutions' ERP assistant, Cortex, has been live since September 2025, but the case study gives no figure.
- Trigent: its client-success page has no agent case study; the "agents" in its insurance cases are insurance agents.
- Leanware, a Clutch sponsor with a $50,000+ minimum: its site blocked our browser, so we read the Internet Archive's copies of its case studies from 21 September 2026. It files five under "Custom AI Agents": two are its own tools, one keeps the client anonymous, and the two with named clients give no figure. One is a crown-grading agent, in production, built for a University of Colorado student's project ("Faculty time saved", with no number); the other is Asana Rebel's UnVape app ("No business or user metrics are public for UnVape").
No step that waits for a person (rule 4)
- SoluLab, which passed rules 1 to 3 with a $25,000+ minimum on Clutch: its Amanbank case names a Libyan bank whose rebuilt app lets customers transfer money, request cards and pay bills by chat and voice. The case has four lines on the bank's support staff: the aim, to free up "human agents to focus on more complex issues", and three results, "a 50% reduction in the workload of human support agents", a voice assistant that "reduced the workload on human agents", and staff left to "focus on more complex and high-value customer issues". Those lines split the work, routine requests to the assistants and complex ones to staff, but none names a point where an assistant stops and waits for a person: the case never says which requests it passes on, or what the voice assistant needs before it moves money. SoluLab's conversational AI page says its assistants escalate "on low confidence or sensitive intents" on most builds, which is its usual practice, not a step this case shows.
Seven questions to ask any agency, and what a good answer sounds like
Send these before the first call and compare the written answers side by side. The last column says where we found a good answer already published, marks it "In part" where we found only some of it, and says so where we found none.
| Question | A good answer | A weak answer | Where we found one published |
|---|---|---|---|
| 1. Which agent you built is in use today, for which client, and what changed, in numbers? | A client's name, the job and one figure, and a reference you can call | Every project is under NDA, or a demo instead of a live system | In part: the three that passed publish the name, the job and the figure; ask them for the reference |
| 2. What may the agent do on its own, and what waits for a person? | A written list of actions, each with the threshold that sends it to a person | Full autonomy from day one, or the rules left until the build | In part: Azumo runs agents "human-in-the-loop above defined thresholds", but doesn't publish the actions or the thresholds |
| 3. How will you test it on our own past cases before launch, and after every change? | A test set built from your cases, a pass mark agreed first, and a rerun on every change | Testing only at the end, by your staff | In part: HyperSense runs 5,000+ automated checks on every code change for one agent, but doesn't say they're built from the client's cases or held to a pass mark agreed first |
| 4. What will it cost to build, and what will it cost each month to run? | Build and running cost quoted apart, with model usage, hosting and upkeep named | An hourly rate and no total | In part: Leanware sets out a setup fee, then a monthly fee covering model costs, hosting and monitoring, and publishes neither figure |
| 5. How much of your work is agents? | A share, and agent projects named | Agents as one line in a long list of services | HyperSense: 40% of the services it lists on Clutch, with Reperks and ZURI named on its own site; Clutch shows every firm's split |
| 6. Who owns the code, prompts and test set, and who runs it after launch? | You own all three at handover, and it runs on your systems or under a named support plan | It runs on the agency's own platform, under a licence | In part: HyperSense says "You own all code, models, and data pipelines" and adds a run-book and training for your operations team, but doesn't name prompts or test sets |
| 7. Did you pay for the place where I found you? | A straight yes or no | A ranking from a list the agency wrote itself | None: no firm we read says; Clutch labels its sponsored listings, so you can check |
Source: the entries above, HyperSense's agent page, Leanware's AI agent page (the Internet Archive's copy of 21 September 2026) and Clutch's directory. Checked 27 September 2026.
What agencies charge to build an agent
Clutch shows each firm's minimum project, so the floor is easy to count. We counted it for all 51 firms on the first page of its AI agent directory:
| Minimum project | Firms |
|---|---|
| $1,000+ | 1 |
| $5,000+ | 4 |
| $10,000+ | 19 |
| $25,000+ | 17 |
| $50,000+ | 6 |
| $75,000+ | 2 |
| $100,000+ | 1 |
| Not disclosed | 1 |
Source: Clutch, AI agent developers, first page (51 firms). Checked 27 September 2026.
On the same page, 21 firms charge $50 to $99 an hour, 13 charge $25 to $49 and 8 charge under $25; 3 charge $100 to $149, 4 charge $150 to $199, and 2 show no rate.
Hourly rates, the same 51 firms on Clutch's first page
The ranking pages that answer the cost question give wide ranges and don't say where they come from. SoluLab's list says a build "typically ranges from $20,000 to $300,000+". DevCom's list puts a proof of concept at $15,000 to $100,000 and production at $25,000 to $500,000 in its FAQ, with no source; DevCom's separate cost guide, which our AI agent cost guide reads, does show its working. Inoxoft gives $25,000 to $75,000 for a narrow single agent, $80,000 to $250,000 with enterprise integrations, and $250,000 and up beyond that. Appinventiv says a build "typically ranges from $40,000 and can exceed $500,000". Our AI agent cost guide works one build through in engineer-weeks and prices the running cost of each finished task.
How to use this list
Pick the group that matches your job and shortlist two or three firms, adding any you met on another list. Send each the seven questions, and keep firms on the shortlist when they can answer the first with a client's name, what the agent does and a measured result, however high or low they sit on a list: of the fifteen firms that reached criterion 3, eleven don't publish all three. Then compare the written answers on approval and testing, not the demo alone.
If you'd like our answers too, read our AI agent development services page for the scope, the approval levels and the four stages, or use the Scope an agent button at the end of this page to send us the seven questions.
FAQs
Frequently asked questions
Who is leading in agentic AI?
If you mean the software, the answer depends on what you already run: Microsoft, Google, Salesforce and ServiceNow all sell agents inside their own systems, and our enterprise AI agents guide compares them. Among the firms that build agents for clients, no single firm clearly leads on published evidence yet: of the 20 we checked, 3 met all five criteria, which include a named client's agent with a result, a step where it waits for a person and a published price floor.
Which AI agent development companies are in the US?
All three that passed show a US headquarters on Clutch: Azumo in San Francisco, with a second office in Rosario, Argentina; EffectiveSoft in San Diego, with offices in Warsaw, Pittsburgh and San José, Costa Rica; and HyperSense in Schaumburg, Illinois. HyperSense's own overview on the same profile, though, describes "a 40+ person team headquartered in Bucharest with a US office in Chicago", so a US address on a profile doesn't mean a US team. On Clutch's first page, 21 of the 51 agent developers show a US city on their card, so ask any firm where your project's engineers will sit.
How long does it take an agency to build an AI agent?
HyperSense sells a one-month sprint for a single agent and reports that Reperks' production system took six weeks. Inoxoft puts a single agent in financial services at four to nine months from discovery to stable production. Ours run a two-week scope that ends in a fixed quote, then a pilot of four to six weeks and a production build of eight to fourteen.
Are agentic AI companies and AI agent development companies the same thing?
On Clutch and the agency lists, both names mean firms that build agents for clients, but three of the top five results for "agentic ai companies" mean something else. Clutch describes its directory of AI agent developers as connecting buyers with "the top AI agent developers and agentic AI companies", and of those top five, SoluLab's list and Inoxoft's video both name firms that build agents for clients. The other three read the words differently: a software vendor's essay, a piece on hiring agent engineers and an investing podcast. If you mean the big software companies whose products include agents, our enterprise AI agents guide compares them.
What are the best agentic AI companies to invest in?
This list can't answer that: every firm in it sells services. The investing podcast on the same results page, from The Motley Fool, is about which big technology companies are ready for agents.
Should I hire an agency or buy an AI agent product?
Consider an existing product first: the right first agent is usually the one inside the system where the job already happens. Hire a firm when the job needs a system no product reaches, when an incorrect action could create a regulated event rather than a service lapse, or when your volume makes a product's unit price cost more than owning the agent. Our AI agents for business guide sets out that test and the products by job.
Written by
Hina Ejaz
Content writer
Hina Ejaz has been a content writer for about seven years, and has written about AI for the last four.
Related reading
All insights- InsightsSelf-hosted AI agents in 2026: 10 you can run on your own servers, and what they send homeWe checked 34 tools for self-hosted AI agents; 10 passed. Where each model runs, where a person approves, what it sends home, and a private model priced.
- InsightsAI agent cost in 2026: what one agent costs to build, run and buyAI agent cost for one finished task: September 2026 model prices, what vendors charge per unit, what a build takes, and the volume at which owning pays.
- InsightsThe best AI agents for business in 2026, by the job they doAI agents for business, sorted by job: what each needs you to run, what it can change, its September 2026 price and unit, and when building one pays.
Put the seven questions to us
Send the job and the systems it touches. An engineer replies within two business days with our answers to all seven, including the ones we can't yet back with a published figure.